Re: [hatari-devel] rev 3689 : splitting cpu cycles above 20

[ Thread Index | Date Index | More lists.tuxfamily.org/hatari-devel Archives ]


On 07/01/2012 20:01, Laurent Sallafranque wrote:
 > I'm not familiar with the falcon, but given the different caches in
the cpu and the different video mode, I think it might be much more
complicated to read the video address in a "sync" way to measure an
instruction reliabily.

No if I always do my tests with cache off.
I know we'll never have a perfect 68030 emulation (too complex).
But timings are important anyway.

 > (I don't even know if some falcon demos are doing "sync" code with
the video beam ?)

I don't know one too (and I've tested a lot).

I just need a precise counter that I can use both in the real falcon and
in hatari to compare some values.


The problem, is that it's a "chicken and egg" issue ; you need a precise videl to measure the cpu, but you also need a precise cpu to be sure videl is correct, else you don't know which one of the 2 is emulated at the correct speed. I don't think timings are not important, I just say it's *really really* complex to measure on the falcon, and that the falcon mode is not ready enough for this. Timings should be the final goal, but in the meantime, I think there're a lot of things that could be progressively fixed in many demos by improving videl (and that don't require exact timings).

There're many roads to reach a good falcon emulation, my opinion is that focusing too much on cycle exact problems for now is not the most efficient way, I'm sure there're many other parts to improve that don't require 100% accuracy and that would improve user experience a lot in demos for example (sound problem as reported by anders in dspmod, videl)

Nicolas



Mail converted by MHonArc 2.6.19+ http://listengine.tuxfamily.org/